Why Are Hispanic and African-American Dropout Rates So High? Magnus Lofstrom

DISCUSSION PAPER SERIES
IZA DP No. 3265
Why Are Hispanic and African-American
Dropout Rates So High?
Magnus Lofstrom
December 2007
Forschungsinstitut
zur Zukunft der Arbeit
Institute for the Study
of Labor
Why Are Hispanic and African-American
Dropout Rates So High?
Magnus Lofstrom
University of Texas at Dallas
and IZA
Discussion Paper No. 3265
December 2007
IZA
P.O. Box 7240
53072 Bonn
Germany
Phone: +49-228-3894-0
Fax: +49-228-3894-180
E-mail: [email protected]
Any opinions expressed here are those of the author(s) and not those of the institute. Research
disseminated by IZA may include views on policy, but the institute itself takes no institutional policy
positions.
The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center
and a place of communication between science, politics and business. IZA is an independent nonprofit
company supported by Deutsche Post World Net. The center is associated with the University of Bonn
and offers a stimulating research environment through its research networks, research support, and
visitors and doctoral programs. IZA engages in (i) original and internationally competitive research in
all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research
results and concepts to the interested public.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be
available directly from the author.
IZA Discussion Paper No. 3265
December 2007
ABSTRACT
Why Are Hispanic and African-American Dropout Rates So High?
The proportion of students who do not graduate from high school is dramatically higher
among the two largest minority groups, Hispanics and African-Americans, compared to nonHispanic whites. In this paper we utilize unique student-level data from the Texas Schools
Microdata Panel (TSMP) in an attempt to determine what factors contribute to the higher
minority dropout rates. We show that poverty is a key contributor. Lack of English proficiency
among Hispanic student is linked to the higher Hispanic dropout probability. Our results also
suggest that neighborhood characteristics may be important in explaining the high AfricanAmerican dropout rates. We also address the issue of the surprisingly low official dropout
rates reported by the Texas Education Agency (TEA) and show that the GED program
explains some of the discrepancy.
JEL Classification:
Keywords:
I21
dropout rate, educational attainment
Corresponding author:
Magnus Lofstrom
University of Texas at Dallas
School of Economic, Political & Policy Sciences
P.O. Box 830688, GR 31
Richardson, TX 75083-0688
USA
E-mail: [email protected]
1. Introduction
A substantial proportion of students do not complete high school, a problem
which is particularly pronounced among the two largest minority groups in the U.S. Hispanics and African-Americans. Greene (2001) report that only 54 percent of Hispanic
students graduate high school and only a slightly higher percentage of AfricanAmericans, 56 percent, complete high school. Amongst white students, the graduation
rate is significantly higher, 78 percent. Hispanic and African-American students in Texas
are also substantially more likely to drop out of school than white students. Using unique
longitudinal student data for the class of 1999 in Texas, we estimate that 40.3 percent of
Hispanic, 38.5 percent of African-American and 26 percent of white public school
students fail to graduate high school by the age of 20.
There are a number of reasons motivating a study of dropout rates, and
particularly the high minority dropout rates. Chances for economic success among
individuals who lack a high school diploma appear to be less today than at any other
point in U.S. Evidence of this is widely reported in both media and academic journals.
For example, Snower (1999) states that “since the mid-1970s in the U.S., the earnings of
the less educated have fallen rapidly behind those of the more educated” (p. 4). The
increasing importance of skills and education in the economy are apparent for economic
outcomes such as employment and earnings. Perreira, Mullan and Lee (2006) report that
in 2000 high school dropouts are almost twice as likely to be unemployed compared to
high school graduates. Data from the 2000 U.S. Census also clearly show that high
school dropouts fare relatively poorly in the labor market. Average annual earnings
among male dropouts between the ages of 25 and 65 was $26,400 while male high school
1
graduates on average earned close to $35,000 per year. Male college graduates in the
same age group earned on average close to $40,000. Among women, the differences
across educational attainment groups are also large. Moreover, the types of jobs available
to high school dropout rarely provide opportunity for significant upward mobility or
benefits such as health insurance. In other words, a relatively certain road to economic
long-term marginality is to not complete, at least, secondary schooling. It is clear that it is
important for policy makers to know what factors and issues are related to poor
educational outcomes, such as dropping out of high school.
This paper has two main objectives. Firstly and primarily, we seek to identify
factors which explain the higher dropout rates among the two largest minority groups in
Texas – Hispanics and African-Americans. Our study is based on unique longitudinal
student level data which include all Texas public school students in the cohort studied,
the class of 1999. Secondly, we generate dropout rates in Texas that arguably more
accurately reflect student high school outcome than official statistics and explain some
reasons for the discrepancies.
2. Previous Research
There is a large body of research focusing on identifying determinants of
educational outcomes and student success (a review of this literature can be found in
Haveman and Wolfe (1995)). Family background, income and parental education are
factors frequently found to affect children’s schooling outcomes. Other determinants are
neighborhood and peer effects, as well as school characteristics. Surprisingly, research
fails to find a consistent relationship between school resources and student achievement
2
(Hanushek, 2006). Research investigating ethnic differences in school outcome, much of
which focus on differences in test scores, have also found the above characteristics to be
factors which help explain minority-white differences in academic outcomes (LavinLoucks, 2006). Next, we discuss relevant economics of education research on ethnic
differences in dropout rates.
The influential recent study by Cameron and Heckman (2001) includes an
analysis of the high dropout rates, specifically the low high school graduation rates,
among Hispanics and African-Americans. They find that family factors (e.g. family
composition, parental education, family income) explain all of the black-white gap in
high school graduation rates and most of the Hispanic-white gap. Cameron and Heckman
(2001) also find that differences between white and minority scholastic ability, as
measured by the Armed Forces Qualification Test (AFQT), play a prominent role in
explaining the white-minority gap. Furthermore, controlling for differences in family
background and AFQT, Hispanics and African-Americans are found to be more likely
than whites to graduate high school. Based on their findings, they conclude that it “is
early differences in resources and not later ones that matter more” (p. 492).
Given the large proportion of immigrants in Texas, particularly Mexican
immigrants, research on high school completion among immigrants is also relevant.
Perreira, Mullan and Lee (2006) find that differences in dropout rates are driven by
differences in human, cultural, school and community capital. They also find that first
generation Hispanic immigrants are less likely to be high school dropouts compared to
their parents but the relative gains in schooling attainment, compared to their parents,
decreases by the second and third generation. Noting that the average Hispanic immigrant
3
has fewer than 9 years of schooling, and has already been out of school for at least one
year by the age of 16, Betts and Lofstrom (2001) suggest that the low levels of
educational attainment among young Hispanic, particularly Mexican, immigrants may be
due to the possibility that they do not “drop in” to high school when they arrive in the
U.S.
In our analysis below we include several of the observable factors included in the
Heckman and Cameron (2001) and the Perreira, Mullan and Lee (2006) studies as well as
school and school district characteristics. Unlike the Cameron and Heckman study, but
like the Perreira, Mullan and Lee study, we are able to identify GED holders and we treat
them as dropouts. Additionally, we are able to control for student’s English ability.
3. Data
We utilize data from the Texas Schools Microdata Panel (TSMP), a student-level
data set made available under special arrangements with the Texas Education Agency
(TEA). These unique data contain information on year, age, grade and school enrolled,
gender, ethnicity, whether the student is economically disadvantaged, whether English is
not the primary home language (ESL), whether the student has been identified as being of
Limited English Proficient (LEP) and whether the student participates in bilingual,
special or gifted education classes. For the high school graduates in our sample, we also
have information on the graduation date. Furthermore, based on age, year and grade
enrolled, we calculate the number of grades a student has been held back.
In constructing our analytic data we begin with a balanced panel of all students
whom we can observe if they are enrolled in Texas public high schools between the ages
4
of 15 and 20. Since the TSMP data currently include student level information from 1990
through 2001, this necessarily means that we restrict the sample to students who belong
to cohorts that would be graduating in the classes from 1993 to 1999. For example,
students who would be expected to graduate in 1993 are 15 years of age in 1990 and they
will not turn 20 until 1995, and so students in this graduating cohort are in our sample
and represent the earliest cohort. Similarly, students expected to graduate from high
school in 1999 are 20 in 2001, and hence students in this graduating cohort are in our
sample and represent the latest available cohort in our data. For the sample of 15 to 20
year-old students who have expected graduation dates from 1993 through 1999, we can
observe, for each year, whether the student was enrolled or graduated from high school.
For our analysis and empirical models we restrict our sample to the latest cohort we can
follow up until they turn 20, i.e. the class of 1999. Nonetheless, we will present dropout
rates for all cohorts in order to show the trend in Texas dropout rates.
Given the structure of the data, we define a student to be a school dropout if we
observe him/her enrolled in a Texas public school at the age of 15 and in some
subsequent year between the age of 15 and 20 this person is not enrolled in a Texas
public school nor have they graduated from high school. Put slightly differently, we use
information on whether the student was enrolled in a Texas public school in a particular
year and/or if the student was reported graduating high school in that year. If the student
is not enrolled and has not graduated high school, but was observed enrolled at age 15,
we define the student to be a school dropout.
5
Caveats – Our Definition of Dropout Rates
Our focus in the paper is students’ secondary educational outcome by the age of
20. We define a person to be a dropout if he/she was observed enrolled in a Texas public
school at the age of 15 and had not graduated high school by the age of 20. Students who
were observed enrolled in a Texas public school at the age of 15 and who either
transferred to a private school, entered home schooling or moved out of state cannot be
tracked in our data and are incorrectly identified as dropouts in the event they actually
graduate high school. Also, students who were observed enrolled at age 15 in a Texas
public school and passed away before graduating high school are also incorrectly
identified as a dropout in our data. Our dropout rates may then be viewed as “upper
bounds”. On the other hand, our dropout rates are underestimated by the fraction of
students who dropout at age 14 or younger.
Our dropout rates are higher than the reported dropout rates by the Texas
Education Agency (TEA) but quite closely correspond to the dropout rates implied by the
high school graduation rates reported by Greene (2001). Greene estimates that 67 percent
of the students in the cohort who should have graduated in 1998 in Texas actually did
graduate, implying a dropout rate of 33 percent. Our calculation yields a dropout rate of
34.6 percent for the 1998 graduation cohort. We will discuss plausible factors explaining
differences between the TEA reported dropout rates and our in the next section.
Data from other Sources – School, District and Local Information
In our empirical models we also incorporate information on school district
characteristics, such as expenditure and revenues per pupil. These data are collected from
6
National Center for Education Statistics’ (NCES) Local Education Agency Finance
Survey for the years 1996 to 2001.1 We also incorporate school characteristics
information, including pupil teacher ratio, enrolment, ethnic/racial student composition
and school location. The school level data are generated from NCES’ Common Core of
Data and are also for the years 1996 to 2001. Lastly, we generate annual local labor
market conditions such as unemployment rate, employment and earnings growth and
average weekly earnings. These data are derived from the Bureau of Labor Statistics’
Quarterly Census of Employment and Wages (QCEW). The local labor market
characteristics are generated at the Metropolitan Statistical Area (MSA) in which the
school district is located. If the school district is located outside a MSA, we utilize the
state average in that particular year.
Descriptive Statistics
We present summary statistics by race/ethnicity for the class of 1999 in Table 1.
The data indicate that there are large differences across groups in student characteristics.
For example, while approximately 17 percent of white students are economically
disadvantaged (defined as eligible for reduced price or free lunch), more than 70 percent
of Hispanics and close to 56 percent of African-Americans are economically
disadvantaged. Not surprisingly, the data also show that Hispanic and Asian students are
more likely to be immigrants, as well as LEP and ESL students than whites or AfricanAmerican students.
1
The period 1996 to 2001 represents the years in which the students in our analytical cohort, the class of
1999, were between the ages of 15 and 20.
7
There are also differences across groups in school and school district
characteristics students attend. The data indicates that white students are more likely to
attend smaller schools with lower pupil teacher ratio. The majority of Hispanic and
African-American students attend schools located in central cities. Students in these two
minority groups also attend schools in district with lower expenditure per pupil. The local
labor market data indicate that among the four groups, Hispanics reside in areas with the
lowest average earnings, earnings growth and employment growth and the highest
unemployment rates. Overall, our data indicate that Hispanics and African-Americans
attend schools, and reside in areas, that are less conducive to academic success than white
and Asian students.
4. Dropout Rates in Texas
Why Are TEA Reported Dropout Rates So Low?
Dropout rates in Texas declined in the 1990s. Figure 1 shows that 38.2 percent of
the class of 1993 did not graduate high school by age 20 while the dropout rate amongst
students in the class of 1999 is 32.7%. TEA also reports a decline in the dropout rate
during this period. However, the TEA reported dropout rates are dramatically lower than
the ones shown in Figure 1. For example, for the latest cohort available in our data, and
the cohort we will focus on in our analysis below, the class of 1999, TEA reports a 9
percent dropout rates (TEA, 2001). An obvious question is why do these dropout rates
differ so much? There are a number of reasons for the discrepancy. For example, TEA
does not define a student who leaves school, without graduating, as a dropout if the
8
student become incarcerated, been expelled for criminal behavior, or has completed all
course work but did not pass the Texas Assessment of Academic Skills (TAAS) exam.
One important reason for the difference the dropout rate is the role of the General
Educational Development (GED) program. The TEA definition of dropouts excludes all
individuals who instead of graduating receive the GED credential. Furthermore, students
who withdraw from school to enroll in an approved alternative program are not counted
as dropouts. This includes work toward completion of the GED certificate, but not
necessarily successfully passing the battery of GED exams.
The distinction between being a high school graduate and holding instead the
GED credential, which is commonly referred to as “high school equivalency” diploma,
would arguably not be important if GED holders did as well in the labor market as high
school graduate and pursued higher education to a similar extent. Research, however,
quite convincingly show that neither of these hold true. Cameron and Heckman (1993)
find that GED holders fare consistently worse than regular high school graduates on any
number of labor market outcomes. Also, research fails to find that the GED increases
earnings among minorities relative to other minority dropouts (Tyler, Murnane, and
Willett, 2000). Regarding higher education, the findings of Lofstrom and Tyler (2005)
suggest that the GED is not a particularly effective route to postsecondary education
relative to staying in school and obtaining a high school diploma. It appears quite clear
that dropouts with the GED are not the equivalent of high school graduates and hence
should not be treated as such in official educational attainment statistics.
To illustrate the importance of the GED in explaining the discrepancy between
our dropout rates and the official TEA dropout rate, we generated the percent of dropouts
9
from the class of 1999 who attempted and passed the GED exam by the age of 20,
obtained from the TSMP data. We present these statistics in Table 2. The data show that
20 percent of dropouts in this cohort obtain the GED credential. If we exclude students
with the GED credential, using our data for the class of 1999, the implied dropout rate is
28 percent, implying that the official dropout rates is understated, at least, by close to 17
percent.2 Table 2 also shows that slightly more than 29 percent of dropouts attempted the
GED by the age of 20. If we exclude from our sample all of these individuals, the implied
dropout rates is 25.6 percent, suggesting that that the official dropout rate is understated
by more than 27 percent. It should be noted that TEA does report the number of students
who graduated in each cohort. For the Class of 1999 cohort, TEA reports that 78.1
percent of this cohort graduated, implying a dropout rate of 21.9 percent (TEA, 2001).
However, this statistic appears to receive substantially less attention than the reported, so
called, dropout rate.
Although these calculations make clear that the GED plays an important role in
explaining the low official dropout rate, they also show that other factors are very
important. For example, as long as a student withdraws from school with the intent to
enroll elsewhere, the student is an “official other leaver” and will not be included in the
statistics used to calculate the official dropout rate. Put differently, these students are not
part of the denominator used to derive the official dropout rate. An indication that this
issue is an important factor in explaining the difference between our dropout rates and the
official TEA dropout rates is the discrepancy in the reported size of the class of 1999.
TEA reports that the class of 1999, grade 7 cohort, includes 240,865 students while the
2
The unadjusted dropout rates shown in Figure 1 is 32.7 percent suggesting that the TEA dropout rates is
understated by, at least, a factor of (32.7– 28)/28 = 0.169.
10
class of 1999 used in this paper contains 298,581 students. Again, the GED, and the
number of students attempting the GED, is a factor explaining the discrepancy. The
number of GED test takers in the class of 1999, according to our data, is 28,365, which
leaves a difference of more than 29,000 students between the TEA reported cohort size
and ours. It appears unlikely that 29,000 students of the class of 1999, about 10 percent of
the cohort, who were enrolled in a Texas public school at the age of 15 enrolled in a
private school, out-of-state school, out-of-country school or became home schooled (and
eventually graduated). Hence, arguably the official dropout rates do not accurately reflect
secondary school attainment among students in Texas.
Differences in Dropout Rates across Groups
The main objective of this paper is to analyze differences in dropout rates and
particularly to identify factors explaining the high dropout rates among Hispanic and
African-American students. Our calculated dropout rates vary substantially across ethnic
groups, shown in Figure 2, and are especially high among Hispanics and AfricanAmericans. According to our data, slightly more than 40 percent of Hispanic students in
the class of 1999 who were enrolled in a Texas public school at the age of 15 did not
graduate high school by the age of 20. The dropout rate is almost as high for AfricanAmericans, 38.5 percent. Both white and Asian students dropout to a significantly lower
extent, 26 and 23.4 percent respectively. In our empirical analysis below we will
investigate the causes of these large differences.
Of obvious interest is also the timing of dropping out of school. That is, at what
age and/or grade are students most likely to drop out? Figure 3 shows unconditional
11
dropout probabilities by age for each of our ethnic/racial groups, i.e. group specific
cumulative density functions (cdf). The figure indicates that relatively few students are
dropouts by the age of 15 and that there are only small differences across groups at this
young age. This is consistent with the findings of Cameron and Heckman (2001), who
note that “few males quit school before age 16” (p. 459). Furthermore, they note that
“this finding is due in part to laws about compulsory schooling attendance and to the lack
of labor market opportunities for people younger than 16” (p .461). The largest difference
in dropout rates in Texas at age 15, of 3 percent, is between Hispanic and Asian students.
However, the dropout gap between these two groups increases with age with a
considerable increase during the year students turn 17. Overall, Figure 3 suggests that the
changes in the cumulative dropout rates over age are quite similar and relatively low for
whites and Asians at one end of the spectrum, with Hispanics and African-Americans on
the other end, displaying alarmingly high dropout rates.
The longitudinal data utilized here can also be used to generate conditional
dropout probabilities. The conditional dropout probability, the hazard rate, is the
probability that a student will dropout by the end of the year, conditional upon being
enrolled during the year. Unlike the unconditional dropout probability, the hazard rate
does not need to be a monotonically increasing function with age. It also has the
advantage of more clearly illustrating the timing of differences in dropout probabilities
across groups. We present the group specific hazard rates in Figure 4.
It appears that the large differences across groups in the dropout rates start at age
16. The figure shows that, conditional on having stayed in school until age 16, Hispanics
and African-Americans are approximately twice as likely as an Asian student to dropout
12
by the end of the year. This roughly holds true for students who continue to be enrolled
the following year, when they turn 17. Approximately 8 percent of Asian, and white,
students drop out during this year while 14 percent of Hispanic students. Among AfricanAmerican students, 15 percent do not continue school further. Not surprisingly, staying
enrolled until age 18, the age at which most students who graduate do so, the conditional
dropout rates decline for all groups. Although the gap in the conditional dropout
probability between Hispanics and whites, as well as the gap between African-Americans
and whites, decrease to 5 percent, these minority groups are more than twice as likely to
leave school, without graduating, at this age than white students. Overall, Figure 4
suggests that the higher dropout rates among Hispanics and whites, compared to whites
and Asians, develop at ages 16 and 17.
Arguably, it is of greater relevance to determine at what grade, as opposed to age,
the dropout differences arise. To shed light on this, we generated group specific
conditional dropout probabilities, hazard rates, by grades. These are presented in Figure
5. The figure quite clearly shows that these differences are particularly generated in the
9th grade and that relatively few students dropout in earlier grades. Approximately 12
percent of African-American students who enrolled in 9th grade did not complete this
grade. Among Hispanic students, an even larger proportion of students did not finish 9th
grade, slightly more than 15 percent. This is in stark contrast to white and Asian students.
For these two groups, 6 percent and 4 percent respectively dropout in 9th grade. Figure 5
also shows quite clearly that these differences decline in subsequent grades, with the
exception of the difference between whites and African-Americans in 12th grade.
Surprisingly, the conditional dropout rates among African-American students who made
13
it all the way to 12th grade is higher than the hazard rate among African-American
students enrolling in the 11th grade. In fact, the conditional dropout probabilities remain
relatively constant among African-American students from grades 9 through 12 while
they decline among Hispanic students in these grades.
The importance of the 9th grade as an explanation for the differences in dropout
rates between the groups can also be illustrated by showing the group specific proportion
of all dropouts who dropped out in 9th grade. About 16 percent of Asian dropouts
dropped out in the 9th grade while about 23 percent of white dropouts enrolled in the 9th
grade without completing it. Among Hispanic and African-American drop outs, 37
percent and 31 percent respectively experienced 9th grade as their last grade.
One plausible explanation for the substantially higher Hispanic and AfricanAmerican 9th grade conditional dropout probabilities, compared to whites, is differential
grade retention across ethnic groups. If a student is progressing through school at a
normal rate, i.e. have not been held back a grade, we would expect him/her to be in the
9th grade by the age of 15. If however a student has been held back one grade, the student
would be expected to be in 9th grade the year he/she turns 16. Importantly, as noted
above, this is also the year in which students, given compulsory student attendance, may
leave school. Furthermore, Cameron and Heckman (2001) report that minority groups are
significantly more likely to have fallen behind in school by age 15 relative to whites.
To investigate whether normal grade progression, or lack thereof, is a key reason
for the large minority-white difference in the 9th grade conditional dropout probability,
we generate the grade specific conditional dropout probabilities by ethnic groups for the
sub-sample of students who were on track at age 15, and were hence enrolled in 9th grade
14
or above, by this age. The hazard rates for this restricted sample are shown in Figure 6.
The figure reveals that although the differences are smaller, large differences remain and
the largest minority-white gaps, specifically, Hispanic/black-white gaps, are still
observed for the 9th grade. It appears that other factors than differences in grade retention
across ethnic groups are the main determinants. Although beyond the scope of this paper,
further research is necessary to determine what specifically causes the increase in the
dropout probability at this point and why minorities have a more difficult time
transferring to high school.
5. Empirical Dropout Probability Model
Table 1 suggests that there are large differences in student, school and district
characteristics across the four ethnic/racial groups. A key objective of this paper is to
determine how these factors relate to the differences in dropout rates across the groups.
The educational outcome analyzed in this paper is whether or not a student of the
class of 1999 who was observed enrolled in a Texas public school at the age of 15 is
observed not completing high school by the age of 20, and hence is defined to be a
dropout. Let the outcome variable yijk equal zero if student i in school j, in school district
k is observed graduating from high school by age 20, and let yijk equal one if the student
is observed dropping out between the ages of 15 and 20.3
We may think of the observed outcome yijk being the result of a latent process in
which a student compares the marginal benefits to the marginal costs of continuing
schooling. In this case, let the continuous latent variable, yijk, which represents the value
3
Given our approach and data, the outcome analyzed is identical to not graduating high school by the age
of 20. One implication, and advantage, of this is that students who leave school but return at a later point
are only defined to be dropouts if they did not graduate by the age of 20.
15
an individual i, in school j, and district k receives from the particular decision in each
school year during the period in which the students is still enrolled and between the ages
of 15 and 20, be specified as:
*
yijk
= α 0 + α1 Hispijk + α 2 Blackijk + α 3 Asianijk + Xijk β + W jk δ + Z k γ + ε ijk
*
yijk = 1( yijk
> 0)
where 1(·) is an indicator function equal to one if the enclosed statement is true, i.e. the
student opted to drop out and zero otherwise. The corresponding, linear dropout
probability model is given by equation 1:
P ⎡⎣ yijk = 1| X, W, Z, Race/Ethnicity ⎤⎦ = α 0 + α1 Hispijk + α 2 Blackijk + α 3 Asianijk
+ Xijk β + W jk δ + Z k γ + ε ijk
(1)
Where Hisp, Black and Asian are indicator variables for Hispanics, AfricanAmerican and Asian students respectively and matrices X, W and Z are defined as;
Xijk
=
Matrix containing student characteristics such as gender, whether
the student is designated as economically disadvantaged, reports
English as a Second Language (ESL) , is designated as Limited
English Proficiency (LEP), indicator variables for whether the
student had been held back a grade or more by age 15, is identified
as an immigrant, and whether he/she participated in special or
gifted education.
Z jk
=
Matrix containing controls for school characteristics, such as pupil
teacher ratio, enrollment, school location, percent of students who
received free lunch and percent of students who are white,
Hispanic, African-American and Asian.
Wk
=
Matrix containing controls for school district characteristics and
local labor market conditions by the MSA the district is located
in. The matrix includes variables for expenditure per pupil and per
16
student federal Title 1 revenues. The local labor market
characteristics included are employment growth, the
unemployment rate, average weekly earnings and earnings growth.
The coefficients of particular interest in the above specification are α1 , α 2 and α 3
which represent the differences in the dropout probability between white and minority
students. An alternative, and far more complex, approach would be to model the dropout
probability as a dynamic process, such as Cameron and Heckman (1998) and Colding
(2006). Given that our empirical approach does not account for possible dynamic
selection bias, the results should be interpreted with some caution.
6. Empirical Results
The model specifications defined above are estimated using a sample of students
who belong to the class of 1999 cohort. Our approach is to start with a parsimonious
model specification and subsequently add controls for factors expected to affect the
dropout decision and group differences in the dropout probability. This will shed light on
what specific factors contribute to the white – minority gaps and to what extent. The
estimated coefficients and t-statistics are shown in Table 3.4
We begin by presenting the Linear Probability Model (LPM) results from our first
specification, Model 1, which only controls for gender and nativity. The objective here is
to investigate whether the high dropout rates among Hispanic students is partially driven
by the higher proportion of immigrants, who may face difficulties adjusting to U.S.
schools. The results show that although immigrant students are significantly more likely
to drop out of school, the higher proportion of foreign born students among Hispanics
4
The standard errors used to derive the t-statistics are White’s robust standard errors. Also, we relax the
assumption that student observations within school are independent, i.e. they are clustered at the school
level.
17
only explain a small proportion of this group’s overall higher dropout rate. Native born
Hispanic students are still about 13 percentage point more likely to drop out than white
students, compared to the unadjusted difference of 14.4 percentage points. Not
surprisingly, given the low proportion of immigrants among African-Americans, the
controls for gender and nativity do not affect the estimated black-white dropout gap,
which remains at 12.5 percentage points.
Minority students are more likely to be economically disadvantaged, as can be
seen in Table 1, a factor likely to have an impact on educational attainment. To assess the
role of poverty on group differences in the drop out probability, we add our measure of
students coming from an economically disadvantaged family, shown as Model 2. The
estimates clearly illustrate the influence of poverty on both the dropout probability and
differences between minority and white students. A student who is economically
disadvantaged is approximately 12 percentage points more likely to drop out of school
than other students. Our estimate also show that, other factors constant, Hispanic students
are approximately 7 percentage point more likely than white students to drop out of
school, while African-American students are roughly 8 percentage points less likely to
complete high school. These results imply that our simple poverty measure explains
almost 1/2 of the Hispanic-white dropout probability difference and more than 1/3 of the
black-white disparity.
Another potentially important factor contributing to the high minority dropout
rates, mainly among Hispanic students, is English proficiency (Valenzuela, 2006). We
add controls for English as a Second Language (ESL) and Limited English Proficiency
(LEP) to our models, Model 3. Both of these variables are found to affect the dropout
18
probability. The results also show that the higher dropout rates among immigrant students
is due to relatively low English proficiency - the estimated immigrant coefficient is now
statistically insignificant. Limited English proficiency is also a contributor to the
Hispanic-white difference in the dropout probability, which decrease to about 4.4
percentage points with the added ESL and LEP variables. The two language factors
explain slightly more than 1/3 of the estimated dropout probability difference between
Hispanic and white students, compared to our Model 2 specification estimate.
We next turn to the role of school characteristics and particularly pupil teacher
ratio and school size, the results are shown as Model 4. Although the school average
pupil teacher ratio is found to affect the dropout probability positively and significantly,
school size is not found to significantly alter students’ decision to leave school before
graduation, in this model specification. Furthermore, these factors do not appear to
explain much of the higher dropout probabilities among Hispanic and African-American
students, relative to white students.
The next model specification, Model 5 is intended to investigate the role of the
location of the school and the ethnic/racial composition of its students, the latter
capturing to some extent peer effects. We show in Table 1 that Hispanic and AfricanAmerican students are much more likely to attend schools that are likely to be located in
the center of large cities than white students. Our data also show that Texas schools are
segregated in the sense that white, Hispanic and African-American students are most
likely to attend schools where their ethnic/racial group is the largest ethnic/racial group.
Clearly, the school composition is a factor of the area, or neighborhood, characteristics,
19
and the effects of composition and location on educational outcome are difficult to
disentangle.
The results for Model 5 indicate that, holding all other factors constant, students
attending schools in central large and mid-size cities are more likely to drop out of school
than students attending schools in any other location. The estimates also indicate that
students attending schools with a higher proportion of economically disadvantaged
students, ceteris paribus, are somewhat less likely to graduate high school. The results
also point to a positive association between the dropout probability and the proportion of
African-American and Asian students. Interestingly, the estimates also imply a negative
correlation between the dropout probability and the proportion of Hispanic students.
Importantly, our estimates show that these location related factors strongly affect the
difference in the dropout probability between African-American and white students.
Once we add these controls to our specification, and holding all factors constant, AfricanAmerican students are no more likely to dropout of school than white students. However,
Hispanic students are still found to be more likely to not complete high school compared
to white students.
Other potentially influential dropout factors are the amount spent per student, i.e.
per pupil expenditure, and revenues received from Title I. The latter funds are
specifically intended to improve the academic achievement of the disadvantaged. It is
possible that differences across school district in the amounts received from Title I, as
well as per student expenditure, may also affect ethnic group differences in the dropout
probability. Our results when we add these two factors to our specification, Model 6,
suggest that higher spending and Title I revenues per student are associated with a lower
20
student dropout probability. However, we do not find any indication that differences in
spending or revenues are factors explaining the high minority dropout rates. Similarly,
when we add our controls for local labor market conditions, shown as Model 7, although
they appear to be linked to the probability of completing high school, they do not seem to
be strong contributors to group differences in the dropout probability.
The most general specification presented so far, Model 7, suggests that the higher
dropout probability among African-American students, relative to white students, can
entirely be explained by our set of observable factors. Observationally similar Hispanic
students are however still predicted to be more likely to drop out of school than white
students, by approximately 3.6 percentage points.
Our last model specification, Model 8, adds an indicator variable for whether the
student had been held back a grade or more by the age of 15. This variable is based on
whether the student was observed enrolled in 8th grade, or lower, the year in which he/she
turned 15. Grade retention is arguably simply a symptom, or predictor, of poor
educational outcomes resulting from academic and personal difficulties experienced by
the student. Viewed from this perspective, a finding that students who were held back a
grade are more likely to drop out of school is neither surprising nor extremely useful
since it is a manifestation of these negative experiences, without indicating which
specific ones caused the student to dropout. Nonetheless, in the specification with the
grade retention variable, Model 8, our estimated minority-white differences in the
dropout probability will tell us what these differences are conditional on grade retention,
i.e. whether a Hispanic student who has been held back a grade is more likely to drop out
than a white student who was held back a grade, all other factors held constant.
21
Our estimates show that differences in grade retention between Hispanic and
white students is a relevant factor. Students who have been held back a grade are
substantially more likely to drop out, by close to 35 percentage points, compared to
students who stay on track. In other words, grade retention is a relatively good predictor
of dropping out of school. The Model 8 estimates suggest that lack of successful grade
progression among Hispanic students result in a higher Hispanic dropout rates than
whites. However, the estimates indicate that Hispanics students who made normal grade
progression and were enrolled in at least 9th grade by age 15 are still more likely to drop
out, by about 3 percentage points, than a white student who had not been held back a
grade at age 15.
Lastly, to determine whether other school characteristics, beyond the ones
specified above, for which we do not have information help explain the difference in the
dropout probability between Hispanic and white students, we also estimated a school
fixed effects specification, shown as Model 9. Note that in this specification, we cannot
include the observable school and school district variables, since they are perfectly
multicolinear with the school fixed effects. The estimated Hispanic-white difference
decreases only marginally to 2.8 percentage points. These results suggest that the school
characteristics included in Models 7 and 8 are the ones most relevant for explaining the
Hispanic-white dropout probability difference. Lastly, the fixed effects results also serve
as a robustness check of our estimated individual student effects. The generally small
differences in the estimates between Model 8 and 9, indicates that the student level
estimates are not driven by unobserved school heterogeneity which may be correlated
with the student characteristics.
22
7. Summary and Conclusions
This paper analyzes the high dropout rates among Hispanic and African-American
students in Texas. Utilizing unique longitudinal student data, we show that Hispanic and
African-American students are about 14 and 12 percentage points, respectively, more
likely to drop out of school than white students. We show that the differences in the
dropout rates appear to develop in 9th grade.
We estimate multivariate dropout probability models to asses what factors affect
the student dropout probability and differences across groups. Factors associated with the
higher minority dropout rates differ somewhat between Hispanics and AfricanAmericans. However, one determinant in common for these minority groups is poverty.
Almost 1/2 of the difference in the dropout probability between Hispanic and white
students stems from the greater prevalence of being economically disadvantaged among
Hispanic students. More than 1/3 of the African-American and white student difference
in the dropout probability is linked to our simple measure of student poverty. We also
find that lack of English proficiency is a key factor in explaining the high Hispanic
dropout rate. Our results suggest that neighborhood characteristics, here simply measured
by the location of the school attended and student race/ethnicity composition of the
school, contribute more to the high African-American dropout probability than school
characteristics such as pupil teacher ratio and expenditure per pupil. This suggests the
need for future research to not only focus on school characteristics but to also determine
the role of neighborhood characteristics on student outcomes.
A secondary objective of the paper is to attempt to explain the discrepancy in
official dropout rates reported by the TEA and the ones derived using appropriate and
23
unique student data such as the one utilized here. We show that the GED program is an
important reason for the differences since dropouts with a GED or students who enroll in
a GED preparation program are not included in the TEA dropout statistics. Given
previous research findings strongly indicating that individuals with the GED credential
are less successful in the labor market and have lower post secondary educational
attainment than high school graduates, we suggest that these dropouts should not be
treated and viewed as high school equivalents.
24
References
Betts, Julian R. and Magnus Lofstrom. (2000) “The Educational Attainment of
Immigrants: Trends and Implications,” in George J. Borjas ed. Issues in the
Economics of Immigration, University of Chicago Press, for National Bureau of
Economic Research, pp. 51-115.
Cameron, Stephen V. and James J. Heckman (1993) "The Nonequivalence of High
School Equivalents." Journal of Labor Economics, 11, no. 1, 1-47.
Cameron, Stephen V. and James J. Heckman (1998) “Life Cycle Schooling and Dynamic
Selection Bias: Models and Evidence for Five Cohorts of American Males”
Journal of Political Economy, 106, 262-333.
Cameron, Stephen V. and James J. Heckman (2001) “The Dynamics of Educational
Attainment for Black, Hispanic, and White Males” Journal of Political Economy,
109, 455-499.
Colding, Bjorg (2006) “A Dynamic Analysis of Educational Progression of Children of
Immigrants” Labour Economics, 13, 479-492.
Greene, Jay P. (2001) “High School Graduation Rates in the United States”, The
Manhattan Institute for Policy Research Civic Report, November.
Hanushek, Eric A. (2006) "School Resources" in Eric A. Hanushek and Finis Welch
(ed.), Handbook of the Economics of Education. Amsterdam: Elsevier.
Haveman, Robert and Barbara Wolfe (1995) “The Determinants of Children's
Attainments: A Review of Methods and Findings” Journal of Economic
Literature, 33, 1829-78.
Lavin-Loucks, Danielle (2006) “The Academic Achivement Gap”, Williams Institute
Research Brief, July.
Lofstrom, Magnus and John H. Tyler, (2005) “Is the GED an Effective Route to
Postsecondary Education?” Paper presented at the 2005 Association for Public
Policy Analysis and Management Fall Research Conference.
Perreira, Krista, Kathleen Mullan Harris and Dohon Lee (2006) “Making it in America:
High School Completion by Immigrant and Native Youth” Demography, 43, 511536.
Snower, Dennis J. (1999) “Causes of Changing Earnings Inequality” IZA Discussion
paper No. 29, January 1999.
25
Texas Education Agency (2001) “Secondary School Completion and Dropouts in Texas
Public Schools 1999-00”, August, Austin, Texas.
Tyler, John H., Richard J. Murnane, and John B. Willett (2000) "Estimating the Labor
Market Signaling Value of the GED." Quarterly Journal of Economics CXV, no.
2, 431-468
Valenzuela, Angela (2006) “The Disappearance of High School English Language
Learners from Texas High Schools” Williams Review.
26
Table 1. Sample Means, Class of 1999, by Race/Ethnicity.
White
Hispanic
AfricanAmerican
Asian
Dropout Rate
0.260
0.403
0.385
0.234
Student Characteristics
Male
Immigrant
Economically Disadvantaged
Limited English Proficiency
English as a Second Language
Gifted Program Participant
Special Education
Held Back a Grade or More
0.515
0.009
0.172
0.004
0.003
0.140
0.132
0.273
0.519
0.116
0.702
0.234
0.183
0.070
0.124
0.486
0.509
0.013
0.558
0.005
0.004
0.093
0.165
0.428
0.506
0.161
0.289
0.176
0.145
0.242
0.030
0.264
School Characteristics
Pupil Teacher Ratio
Total Enrolment
Enrolment Less than 700
Enrolment 700 - 1,400
Enrolment 1,400 - 1,900
Enrolment 1,900 - 2,500
Enrolment Greater than 2,500
Percent Free Lunch
Percent White
Percent Hispanic
Percent African-American
Percent Asian
School Location
Central City, Large City (250K+)
Central City, Mid-size City (<250K)
Urban fringe, Large city
Urban fringe, Mid-Size city
Large town
Small town
Rural
14.4
1,561
0.241
0.198
0.176
0.200
0.184
17.8
66.9
19.7
10.2
2.9
15.2
1,685
0.149
0.195
0.229
0.283
0.144
40.8
26.7
61.8
9.5
1.8
15.5
1,636
0.144
0.264
0.229
0.228
0.135
29.4
34.3
23.2
39.1
3.2
16.0
2,318
0.047
0.086
0.158
0.330
0.379
17.7
50.6
23.3
16.4
9.5
0.164
0.139
0.327
0.052
0.010
0.112
0.196
0.388
0.202
0.151
0.057
0.011
0.100
0.090
0.461
0.173
0.195
0.021
0.012
0.065
0.074
0.351
0.104
0.441
0.018
0.002
0.026
0.057
School District Characteristics
Expenditure per Pupil
Title 1 Revenue per Pupil
6,526
103
6,326
216
6,170
157
6,463
89
Local Labor Market Conditions
Employment Growth (%)
Unemployment Rate (%)
Average Weekly Earnings ($)
Earnings Growth (%)
1.8
4.4
639
4.6
1.9
6.2
576
4.3
1.8
4.4
655
4.7
1.9
4.3
680
4.7
145,365
103,809
40,887
7,830
Sample Size
27
Table 2. Number and Percent of Dropouts Who Passed and Attempted the GED, Class of
1999, by Race/Ethnicity.
White
Hispanic
African-American
Asian
Native American
All Students
Total
Number of
Students
145,365
103,809
40,887
7,830
690
Total
Number of
Dropouts
37,785
41,886
15,751
1,834
293
298,581
97,549
Number of
Percent of
Dropouts
Dropouts
Who Passed GED
10,806
28.6
6,230
14.9
2,196
13.9
260
14.2
60
20.5
19,552
28
20.0
Number of
Percent of
Dropouts
Dropouts
Who Attempted GED
13,830
36.6
10,223
24.4
3,882
24.6
348
19.0
82
28.0
28,365
29.1
Table 3. Linear Probability Models of the Probability of Dropping Out of School, Class of 1999.
Variable
Hispanic
African American
Asian
Native-American
Student Characteristics
Immigrant
Male
Economically Disadvantaged
Limited English Proficiency
English as a Second Language
Gifted Program Participant
Special Education
School Characteristics
Pupil Teacher Ratio
Total Enrolment
Enrolment 700 - 1,400
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Model 7
Model 8
Model 9
0.131
(16.78)
0.125
(12.99)
-0.042
(4.53)
0.162
(7.75)
0.069
(9.03)
0.079
(8.29)
-0.054
(6.39)
0.142
(7.02)
0.044
(5.99)
0.083
(8.73)
-0.071
(8.47)
0.142
(7.00)
0.043
(6.20)
0.070
(7.05)
-0.021
(2.27)
0.126
(6.41)
0.039
(8.05)
-0.015
(2.82)
-0.053
(7.08)
0.111
(5.75)
0.039
(8.25)
-0.014
(2.52)
-0.052
(6.92)
0.105
(5.45)
0.036
(7.62)
-0.009
(1.92)
-0.053
(7.80)
0.087
(4.64)
0.031
(6.62)
-0.009
(1.79)
-0.051
(7.48)
0.086
(4.77)
0.028
(11.14)
-0.005
(1.55)
-0.054
(11.14)
0.104
(5.98)
0.111
(6.38)
0.071
(29.40)
0.095
(5.77)
0.070
(28.75)
0.121
(20.22)
0.011
(0.83)
0.068
(28.66)
0.109
(17.82)
0.073
(3.83)
0.130
(5.90)
0.026
(2.31)
0.056
(26.77)
0.082
(14.24)
0.056
(3.62)
0.137
(7.44)
-0.227
(30.99)
0.074
(14.54)
0.013
(1.08)
0.055
(26.57)
0.076
(15.04)
0.053
(3.70)
0.136
(8.19)
-0.240
(25.77)
0.076
(15.76)
0.017
(1.69)
0.056
(26.66)
0.076
(15.15)
0.046
(2.70)
0.142
(7.55)
-0.232
(24.61)
0.078
(16.20)
0.041
(2.58)
0.052
(23.53)
0.071
(13.32)
0.048
(1.95)
0.128
(4.78)
-0.181
(19.51)
0.074
(11.54)
0.037
(2.35)
0.048
(21.91)
0.066
(12.46)
0.042
(1.79)
0.120
(4.68)
-0.175
(19.60)
0.062
(9.76)
-0.006
(1.33)
0.046
(28.76)
0.077
(37.08)
0.046
(7.04)
0.130
(17.90)
-0.234
(117.60)
0.060
(22.59)
0.008
(3.12)
-0.005
(0.21)
0.024
(1.23)
0.006
(2.85)
-0.033
(1.52)
-0.007
(0.41)
0.006
(2.77)
-0.044
(2.05)
-0.023
(1.21)
0.005
(2.43)
-0.052
(2.53)
-0.001
(0.07)
0.005
(2.46)
-0.047
(2.30)
-0.003
(0.18)
29
Table 3 (Continued)
Enrolment 1,400 - 1,900
Enrolment 1,900 - 2,500
Enrolment Greater than 2,500
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Model 7
Model 8
-0.013
(0.40)
-0.073
(1.77)
-0.070
(1.15)
-0.059
(1.87)
-0.103
(2.62)
-0.084
(1.46)
0.002
(3.40)
-0.0012
(3.09)
0.001
(2.74)
0.003
(2.00)
-0.060
(1.92)
-0.100
(2.56)
-0.081
(1.42)
0.002
(4.31)
-0.001
(1.45)
0.001
(2.95)
0.003
(2.03)
-0.022
(0.65)
-0.073
(1.85)
-0.025
(0.41)
0.001
(1.58)
-0.002
(3.04)
0.003
(5.55)
0.008
(4.12)
-0.015
(0.45)
-0.062
(1.56)
-0.016
(0.27)
0.001
(0.83)
-0.002
(2.74)
0.003
(5.81)
0.007
(3.90)
0.191
(10.93)
0.155
(10.34)
0.140
(10.20)
0.032
(1.78)
0.080
(2.01)
0.033
(3.05)
0.162
(8.94)
0.129
(7.65)
0.117
(8.33)
0.021
(1.05)
0.044
(1.01)
0.025
(1.95)
0.150
(5.18)
0.015
(0.38)
0.140
(6.61)
-0.173
(4.25)
0.045
(0.64)
0.033
(1.94)
0.136
(4.76)
0.003
(0.07)
0.130
(6.28)
-0.173
(4.36)
0.040
(0.56)
0.032
(1.94)
Percent Free Lunch
Percent Hispanic
Percent African-American
Percent Asian
School Location
Central City, Large City (250K+)
Central City, Mid-size City (<250K)
Urban fringe, Large city
Urban fringe, Mid-Size city
Large town
Small town
30
Model 9
Table 3 (Continued)
Model 1
Model 2
Model 3
Model 4
Model 5
School District Characteristics
Expenditure per Pupil
Title 1 Revenue per Pupil
Model 6
Model 7
Model 8
-0.037
(2.96)
-0.342
(4.39)
-0.033
(2.90)
-0.485
(4.24)
-0.032
(2.89)
-0.447
(3.96)
0.024
(5.27)
-0.002
(13.55)
0.064
(8.48)
0.064
(8.48)
0.023
(5.32)
-0.002
(13.88)
0.064
(8.63)
0.064
(8.63)
Local Labor Market Conditions
Unemployment Rate
Weekly Earnings
Employment Growth
Weekly Earnings Growth
Grade Retention
Held Back a Grade at Age 15
Number of Students
R-squared
0.031
0.043
0.052
0.091
31
298,581
0.113
0.128
0.259
Model 9
0.349
(37.66)
0.302
(61.04)
0.275
0.279
Figure 1
Dropout Rates by Age 20, Texas,
Graduating Class of 1993-1999
50%
38.2%
40%
38.1%
37.7%
36.8%
35.8%
34.6%
32.7%
30%
20%
10%
0%
1993
1994
1995
1996
1997
1998
1999
Figure 2
Dropout Rates by Age 20, Texas, Graduating Class of 1999
by Race/Ethnicity
50%
40.3%
38.5%
40%
30%
26.0%
23.4%
20%
10%
0%
Non-Hispanic White
Hispanic
African-American
32
Asian
Figure 3
Dropout Probabilities by Age (CDF),
Texas, Class of 1999 by Race/Ethnicity
45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
Age 15
Age 16
White
Age 17
Hispanic
Age 18
African-American
Age 19
Age 20
Asian
Figure 4
Conditional Dropout Probabilities by Age (Hazard Rates),
Texas, Class of 1999 by Race/Ethnicity
35%
30%
25%
20%
15%
10%
5%
0%
Age 15
Age 16
White
Age 17
Hispanic
Age 18
African-American
33
Age 19
Asian
Age 20
Figure 5
Conditional Dropout Probabilities by Grade (Hazard Rates),
Texas, Class of 1999 by Race/Ethnicity
16%
14%
12%
10%
8%
6%
4%
2%
0%
Grade 7
Grade 8
White
Grade 9
Hispanic
Grade 10
African-American
Grade 11
Grade 12
Asian
Figure 6
Conditional Dropout Probabilities by Grade (Hazard Rates),
Texas, Class of 1999 by Race/Ethnicity
Normal Grade Advancement at Age 15
12%
10%
8%
6%
4%
2%
0%
Grade 9
Grade 10
White
Grade 11
Hispanic
African-American
34
Grade 12
Asian
35